Skip to content

Conversation

@SzymonOzog
Copy link
Contributor

Currently when using CUDA with TCP it assumes that we allocated a new buffer in line 152, it's not always the case and we can free memory that doesn't belong to us causing a segfault later

@stmatengss
Copy link
Collaborator

Currently when using CUDA with TCP it assumes that we allocated a new buffer in line 152, it's not always the case and we can free memory that doesn't belong to us causing a segfault later

That's correct. Thank you for the reminder. I believe this is the root cause of issue sgl-project/sglang#7211 (comment).

@stmatengss stmatengss merged commit bdc7ffb into kvcache-ai:main Aug 13, 2025
11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants